Day 2: Unsupervised Learning Examples

We know that ChatGPT was trained using a form of unsupervised learning. But where else can we see this ML task in action?

According to “An Introduction to Statistical Learning”: this task can be used to find patterns in cancer research or how people shop online. You can use those patterns to learn about the disease or make decisions to increase how much a customer spends on your website.

For example, let’s say TikTok can classify you into a group of users called Group A because you like watching dance videos and getting book recommendations. If the majority of users in Group A also like Taylor Swift, then TikTok knows to recommend you Taylor Swift content. Without any prior knowledge, the unsupervised learning task would determine what types of content categorizes Group A and which users should be placed in it. If you decide you scroll past the Taylor Swift content, the task would then re-evaluate which group to place you in or which patterns are relevant for that Group.

Now this is a slightly different example, but unsupervised learning is also used in detecting damage in structures like bridges and highrise buildings. There’s a lot more data on normal structures than damaged ones, so unsupervised learning works well in this case, because it can just use the data on normal structures and spot any changes and deviations from it to create those patterns. I’ll link the research paper for this topic if you want to learn more.

Sources:

• “Chapter 12: Unsupervised Learning” from “An Introduction to Statistical Learning”
• “Unsupervised Learning Methods for Data-Driven Vibration-Based Structural Health Monitoring: A Review”